This is the ILWIMI data portal web site.  Data posted here are spatial data layers for use in the EPA STAR Biological Classification project that focuses on how land use change impacts ecosystem management across different statewide planning units.  Projections of future land use are stored here, along with the data used to construct the model.  You can go to the LTM (Land Transformation Model) web site to learn more about how this model works.

Data are being distributed as zipped files of ArcGIS GRID files.  You can also find other datasets here such as the population projection spreadsheets used to estimate new amounts of urban into the future.

If you are not part of the ILWIMI project and are familiar with GIS and GRID files, feel free to download.   Just send me an email to let me know that you are using the data (bpijanow(at)purdue.edu). There are many caveats to the use and misuse of these data so “user beware”.  Some of the model output have good calibration data and I have a good sense of how well the model is performing; others do not.  These vary in quality by state.  I should also say that we had to invent new procedures to get these projections together, some of which have not been published.  We are in the process of documenting these (Dec 2006) but it is likely to take a while.

The files are organized by state and version number.  Note that version 9 represents the final cut for each of the states.  I’m also distributing version 8 which does not include new forests.  Compare version 8 and 9 to test how ecological response models are impacted by certain land use/cover transitions occurring in the landscape.

Note to ILWIMI project scientists: the files marked with an asterisk (*) are preferred files to run hydrologic and fish simulation models.  The files are also in different projections.  We modeled in an equal area projection (Albers) and then reprojected back to the original so that these can be used in each of the states. 

PROJECTWIDE BASE FILES

The files that contain the crosswalk land use/cover codes for all states and the ILWIMI project are located here. 

                        Crosswalk excel spreadsheet

An ArcGIS layer file that contains the land use/cover codes, names and color maps are located here as well.

                        A layer file (load and apply in Symbology)

Some graphics that you might be able to use in presentations.  I’ll continue to update this file.

                        Graphics file here.

WISCONSIN

Basic Method.  We used two sets of simulations to construct the WI projections.  We used the SW Wisconsin Planning Commissions land use database to train on urban change.  The urban change model was applied to all urban areas in the state.  The aforestation patterns were derived from a central-northern Michigan training of new forest growth and the amounts of new forests were calculated from a MI land use change analysis.

Base Maps.

The map from which all projections are made can be found here:

                        Wisconsin Base Map

The source of the data is the Lillesand et al. wiscland database.   Google wiscland and you will be able to acquire the original data.  We had to reclass and resample the data to have it conform with the ILIWIMI and Aquatic GAP projects.

I will make available the population projections and urban calculation spreadsheet here in the near future.

LTM Output.

Version 9. Urban and forest changes are included as part of the projections.  New forests are grown at 1% per year from the base amount.  Urban change is a function of population and the amount of urban in two subclasses, residential and commercial.  We fixed the original amount from base (wiscland) and used a very simple cellular automata model that selects urban subclass according to nearest neighbor and proportions.  New forests were selected on the basis of nearest neighbor.

                        Version 9 results – with highways (*)

                        Version 9 results – without highways

Version 8.  Only new urban is added to the projections and forests remain static in the model.  Over the course of the 30 year projections, forests decline slightly due to urbanization.  We also combines files with highways and without strong highways.  Use this version and compare results of any simulation with version 9 (new forests added).

ILLINOIS

Basic Method. We did not have time series data for IL to build the IL model.  Instead, we used central IN data from Jeff Wilson at IUPUI to train on urban change in an agricultural dominated setting and then apply these results to areas outside of Chicago and East St. Louis .  We used the NEIPC land use data to build the Chicago area simulation.   Our analysis of IN land use change databases suggested that aforestation was not occurring in agricultural intensity settings so we decided to not change agriculture and forest amounts in the future.  In other words, IL we are anticipating that no new forest will be added to IL in the future; this differs from MI and WI where we believe that aforestation, especially along the agriculturally margin areas of both states (central) will experience aforestation that is approximately 68% of the rate of urban expansion.

Base Maps.

The map from which all projections are made can be found here:

                        Illinois Base Map

The source of the data is from the Illinois Natural History Survey and is a Landsat TM derived database.  We had to reclass and resample the data to have it conform with the ILWIMI and Aquatic GAP projects.  However, a version of the output is also available using the original classes.

LTM Output.

Version 9. We add new urban at the same rate as in WI and MI and urban classes are separated into the 11 and 12 subclass.  No new forests are added reflecting the intensive agriculture occurring in the area.  A parallel analysis of central IN land use/cover data over 20 years shows no increase in forests (a small loss occurred).

                        Version 9 results are here (*)

                        The projection information for these files is located here.

Version 8. We added new urban using the methods described above EXCEPT for the fact that we did not apply the CA routine to separate out the suburban classes.  We worked on IL Spring of 2006 and did not develop the CA tool until summer 2006.  All new urban is coded as 10.  We reapplied the the CA model (a modified one) that reclasses the 10s into 11 (commercial) and 12 (residential).

                        Version 8 results are here.

MICHIGAN

            Base Maps.

The map from which all projections are made can be found here:

                        Michigan Base Map

The source of the data is from the Michigan Department of Natural Resources IFMAP project.  Google MDNR and IFMAP and you should be able to download the data in two sections (UP and LP).  We had to stitch them together and then reclass and resample.

LTM Output.

Version 9. This version is compatible with the other version 9 products for IL and WI.  We added new forests and new urban to the map.  New forests are grown at the same rate applied to WI.  The aforestation routine was applied to the UP and LP separately and then merged.

                        Version 9 results are here. (*)

                        Projection information for these files are found here.

Version 8. We used data from 17 counties of MIRIS land use change (1978 to around 1998) to train on urban expansion and aforestation patterns across the state. We added new urban using the methods described above which are placed in urban subclasses.

                        Version 8 results are here

HOW WE DEVELOPED THE MODELS

The modeling contained here reflects several stages of work.  First, we conducted an analysis of land use/cover changes occurring across the Great Lakes Basin using high quality, aerial photography acquired from several research projects and from a variety of municipalities in the basin willing to share their data.  Change analysis concentrated on how urban rates varied with population change, how forests and other types of transitions varies across the region and how variable the rates of change occurred across a 20-30 time period.  In general, we found that urban increased by 4.35 times the rate of population increase between 1980 to present, that forests grew at a 1% rate of increase from base forest quantity during each 5 year time period in MI and WI.  No new forests occurred in IL or IN.

We also used base land/use cover datasets developed by each of the states. These are described above.  Population projections by county were acquired from each of the state’s demographic offices and then compared against US statewide projections and county estimates and against some linear models that we developed.  We selected county projections that seemed to strike a moderate rate of increase (a “midpoint” of the three) although a few counties were judged to be outliers and these adjusted by our own model to conform to neighboring counties.  For example, we identified about 10 counties that appeared to be under projected in terms of population growth based on high growing neighbors and these were adjusted upward consistently with their neighbors.

We also conducted an analysis of the relationship to urban growth and population change in counties that experienced a decline in population during the time represented by the land use data.  In general, we found that these counties (mostly in MI) experienced a 1% rate of urban growth every 5 years and this value was applied across the entire study region.

The general outcome of having the entire area double in urban during the 2000-2030 time period is very consistent with our analysis.  Higher rates of growth (e.g., around 8.7x population growth) are anticipated if urbanization is more close to the more recent (1995-2005) trends as opposed to what occurred in the mid to early 1980s (2.6x population growth).

We used the LTM to build “training sets” where 10-20 year land use change data created neural network weights that were then applied to (generally) a different set of data to project forward.  Figure 2 below shows the sources of training set data and areas that they were applied to create the forecast models.  Seven different training and testing (assessing goodness of fit and then forecasting) sets were conducted.  Training and testing set #1 was conducted using land use/cover data from Jeff Wilson (IUPUI) for central Indiana (1983 to 2001) and then the network file was applied to an identical set of GIS processed driving variables (see Pijanowski et al. 2005 in IJGIS).  Analysis of central Indiana land use change data suggested that no new forests should be introduced in Illinois .  Base Illinois data are from the Illinois Natural History Survey 2000 land use/cover database.  The second training set was from central, western and northwestern Lower Peninsula of Michigan.  This training set was used to forecast all of Michigan (#2 and #4) except Southeast Michigan .  It was also used to forecast all of WI (#3) except for the Milwaukee metropolitan area.  Data from the Southeast Wisconsin Regional Planning was used to forecast itself (#5) and data from Southeast Council of Governments was used to also forecast itself (#6).  Finally, data from Northeastern Illinois was used to forecast itself (#7).  All data were converted to the USGS GAP land use cover coding system that is found here (cross walk land use table).

Developed by Bryan C. Pijanowski

Department of Forestry and Natural Resources

Purdue University

West Lafayette , Indiana 47906

bpijanow(@)purdue.edu

765-496-2215